NVIDIA has released GR00T-H-N1.7, the first commercially licensed AI foundation model built specifically for surgical and healthcare robotics, giving robot makers a pre-trained starting point for skills like suturing, tissue manipulation and endoscopy.
A commercial license changes the game
The 3-billion-parameter vision-language-action model upgrades the earlier research-only GR00T-H that shipped in March 2026, adding a new backbone and training data spanning 20 robot platforms and more than 50 institutions. Crucially, it ships under the NVIDIA Open Model License, making production deployments legal for the first time. NVIDIA is explicit that the model is not cleared for clinical deployment or patient care; it is a development substrate for R&D teams.
Solving the hardware fragmentation problem
Surgical robots vary enormously in geometry and instrumentation, which has long made shared AI policies impractical. GR00T-H-N1.7 tackles this with per-embodiment MLP projectors, small learnable networks that translate each robot's specific measurements into a shared, normalized action space. That lets a single foundation model adapt across very different platforms rather than starting from scratch each time.
Who benefits first
CMR Surgical, maker of the Versius system, is both a leading data contributor and an immediate beneficiary, gaining a simulation stack to test new robot behaviours without occupying operating-room time. The release sits alongside NVIDIA's broader physical-AI push, which also produced the Isaac GR00T reference humanoid and the ENPIRE autonomous research framework. It also lands amid intense surgical-robotics activity, including CMR Surgical's U.S. Versius launch.
Reporting based on coverage from NVIDIA Newsroom, TechTimes and 2 Minute Medicine.